Networks / Pajek
Package for Large Network Analysis

How to
Some hints on
working with extremely large networks in Pajek


Extremely large networks

The current limit on the size of Pajek network is almost 10 million vertices, the number of lines is limited by available (virtual) memory (limited by physical constraints to 2 G). The use of virtual memory substantially reduces the performance of Pajek. Use Windows Task Manager / Performance to monitor the usage of memory.

Windows Task Manager

We consider a network as extremely large if it occupies more than half of the available physical memory on the computer. An example of such network is the US patents network (3774768 vertices and 16522438 arcs) from NBER.

The simplest solution to deal with extremely large networks is to increase, if still possible, the physical memory of your computer.

To support work with such networks Pajek provides some special options:

Input without vertex labels

In Options / Read/Write we can switch OFF the option Read vertices labels. The network will be read without labels, thus saving a lot of memory.

Labels OFF

Adding labels to small derived subnetworks

You can add the labels later to a subnetwork using the option Net / Transform / Add / Vertices Labels from File

Adding labels

To speed up reading of network or to be able to select the labels, you may consider to split the network data to file without labels and file(s) of labels. You can even have several files of labels (short/long, different languages).

In-place operations

For larger networks Pajek asks the user whether to store the resulting network as a new network or to transform the current network (thus saving memory).

In Place

Disposing of data

For all its data structures Pajek provides in the File menu the disposing option allowing to the user to dispose the data that are not needed, thus freeing the corresponding memory.

Dispose

How to?; Pajek; Vlado/Networks